Marking this as verified for Noble - My test of the new functionality worked as expected on Noble on DGX A100, and I observed no regressions in my control tests:
Control test 1 (no custom X-PciMmio64Mb): 1. Started a VM with 4 GPUs passed-through without specifying a custom opt/ovmf/X-PciMmio64Mb or pci=realloc pci=nocrs. Boot time post-ram-alloc was ~3:00. Installed `nvidia-driver-550-server` in guest and ran `sudo nvidia-smi` - worked as expected Control test 2 (custom X-PciMmio64Mb > required GPU aperture size): 1. Started a VM with 4 GPUs passed-through with opt/ovmf/X-PciMmio64Mb = 1262144, without pci=realloc pci=nocrs. Boot time post-ram-alloc was ~3:00. Installed `nvidia-driver-550-server` in guest and ran `sudo nvidia-smi` - worked as expected New functionality test (custom X-PciMmio64Mb < required GPU aperture size): 1. Started a VM with 4 GPUs passed-through with opt/ovmf/X-PciMmio64Mb = 1024Mb and pci=realloc pci=nocrs set in guest. Boot time post-ram-alloc was 0:16 2. Installed `nvidia-driver-550-server` in guest and ran `sudo nvidia-smi` - worked as expected I'll also test this on DGX H100 early next week as a final sanity check before it gets promoted (since it was in use tonight) as a sanity check and will update this bug report if I see anything strange/unexpected, but the key behavior here has always been the same between A100 and H100 in my tests so far. ** Tags removed: verification-needed verification-needed-noble ** Tags added: verification-done verification-done-noble -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/2101903 Title: Backport "OvmfPkg: Use user-specified opt/ovmf/X-PciMmio64Mb value unconditionally" to Noble To manage notifications about this bug go to: https://bugs.launchpad.net/edk2/+bug/2101903/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
